A kinase chemically phosphorylates others proteins. You could precisely identify that your protein is a kinase by determining if your protein phosphorylates another protein.
You're asking how to do this bioinformatically (which would not be definitive), but I would look at the protein composition (amino acid sequence), the estimated folding structure, and any additional tertiary indications and see how homologous these characters are in relation to other kinase proteins in databases, such as GO. Does your protein have motifs in common with other proteins that have been fully characterized to be kinases?
You shouldn't use GO terms to define protein kinases. There are more sophisticated ways to make these predictions, such as sequence scanning, on a domain level. A solution would be to use HMMER, combined with the kinomer library (this is similar to what PFAM does), except it will predict on a kinase family level.
Where seqs.fasta is a file containing the protein sequences you want to scan in fasta format, allPK.hmm is the HMM library downloaded from kinomer, and predictions.txt is the file containing the resulting output.
Alternatively, if you want to query a single gene or list of genes against these categories you can use the 'Search Categories' function of DGIdb and enter your gene(s) and filter against which sources you trust as a definition of kinases.
Kinases in DGIdb are defined in various ways by different sources (including the Gene Ontology). A recent "reliable" list would be the dGene list which is based on review of the literature. See the sources and their corresponding publications for more details of how they defined kinases.